Re: [Xen-devel] [RFC PATCH 11/13] cpufreq: add xen-cpufreq driver

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 10.10.14 at 11:54, <stefano.stabellini@xxxxxxxxxxxxx> wrote:
> On Fri, 10 Oct 2014, Jan Beulich wrote:
>> >>> On 10.10.14 at 11:39, <stefano.stabellini@xxxxxxxxxxxxx> wrote:
>> > On Fri, 10 Oct 2014, Jan Beulich wrote:
>> >> >>> On 08.10.14 at 15:51, <stefano.stabellini@xxxxxxxxxxxxx> wrote:
>> >> > On Tue, 7 Oct 2014, Jan Beulich wrote:
>> >> >> >>> On 07.10.14 at 16:19, <oleksandr.dmytryshyn@xxxxxxxxxxxxxxx> wrote:
>> >> >> > This driver uses Dom0 to change frequencies on CPUs
>> >> >> 
>> >> >> In which case its name should be dom0-cpufreq. But I very much
>> >> >> question a model where it is not the hypervisor controlling aspects
>> >> >> of physical CPUs.
>> >> > 
>> >> > Me too, however we had a long discussion (see
>> >> > CAN58jiuVOJMX2-U=Odqrjtr1agNPSRnS1nXURYitc8AC8FgH3g@xxxxxxxxxxxxxx) and
>> >> > Oleksandr and others thought that it would require too much drivers code
>> >> > to change cpu frequency directly in Xen on ARM
>> >> > (CAH_mUMNQLHeOWFC_SNB_8BjBz9rOs=moYOUFFhtOXo_WPZTa7w@xxxxxxxxxxxxxx and
>> >> > CAH_mUMM0iSCcCcYz1B13p5YdS+wvgBOMVJh-871v0Ga0f1bH8Q@xxxxxxxxxxxxxx).
>> >> > 
>> >> > They might be right.
>> >> 
>> >> I continue to disagree, despite understanding some of the points
>> >> they make. ACPI has obtained I2C resource handling a while ago,
>> >> so rather than keeping I2C out of Xen we should rather consider
>> >> updating the few ACPI pieces we have to be capable of handling
>> >> that resource type. Of course that's only useful if systems also
>> >> describe their frequency change procedure properly in ACPI (but
>> >> if they don't do now, I think with ARM moving the ACPI way we
>> >> can at least expect that to come).
>> > 
>> > ACPI is not coming to ARM in general, only to (some?) ARM servers.
>> > Device Tree is alive and well and we have to keep supporting it for all
>> > the other systems.
>> 
>> Okay, but that doesn't change the picture: I2C support is going to
>> be useful (needed?) for ACPI sooner or later, so having something
>> (hopefully minimalistic) in the hypervisor doesn't sound like all that
>> bad an idea, and could then be used outside of ACPI too for ARM.
> 
> Why is I2C needed for ACPI? I thought that all the freq changing ops
> would be hidden behind AML methods.

But the AML methods specify certain resources to be read/written.
Apart from MMIO and I/O ports (the latter on x86 only), this could
now also be an I2C access aiui.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel